Human study PLS-00109

Identifying responders to elamipretide in Barth syndrome: Hierarchical clustering for time series data.

Randomized controlled trial · 12 participants enrolled

Version 1 · published November 7, 2026 · evidence through November 7, 2026 · research snapshot 2026.1.1 · methodology ev-1 · Automated assessment by AI (claude-opus-5-5 and claude-sonnet-5-5), not reviewed by a person.

Only the abstract was available. Essential methods may be missing, which limits what can be judged.

Reports of this study

Every publication, registration and abstract describing this one study. Participants are counted once.

Identifying responders to elamipretide in Barth syndrome: Hierarchical clustering for time series data.primary · Orphanet journal of rare diseases · 2023-04-11 · PMID 37041653 · DOI

Design and participants

Design: Randomized controlled trial
Population: Patients with Barth syndrome enrolled in a randomized, double-blind, placebo-controlled crossover trial of elamipretide.
Conditions: Barth syndrome
Planned enrollment (registry): Not reported
Enrolled: 12
Randomized: Not reported
Completed follow-up: Not reported
Funding: Research Support, Non-U.S. Gov't (per PubMed publication type); funder not named.
Authors’ conflicts: Not reported in the record.

Intervention and comparator

What each group received. Related peptides, analogs and combinations are never treated as the compound itself.

Elamipretide (intervention) · elamipretide · Elamipretide given in the crossover trial; details not reported in the abstract · Not reported assigned
Placebo (comparator) · Placebo period of the crossover trial · placebo · Not reported assigned

What the authors reported

As stated in the sources. This is the authors’ account, not Peptide Library’s assessment.

Accuracy of clustering patients by functional status from wearable physiological data

post hoc outcome · surrogate · Agglomerative hierarchical clustering of wearable-device time series (heart rate, respiratory rate, activity, posture) against median-split functional scores (6MWT, PROMIS fatigue, SWAY balance, BTHS-SA fatigue, dynamometry, 5XSST, MLCL:CL)

—: narrative — Clustering by functional status reached accuracies of 60-93% across the functional scores
—: proportion 93 — Accuracy of 93% for 6MWT functional status
—: proportion 87 — Accuracy of 87% for PROMIS fatigue functional status
—: proportion 80 — Accuracy of 80% for SWAY balance functional status

Accuracy of clustering responders vs non-responders to elamipretide

post hoc outcome · surrogate · Agglomerative hierarchical clustering of wearable physiological data against median-split best and worst response to elamipretide on the functional scores

Elamipretide: proportion 100 — All response-to-treatment clustering models classified responders vs non-responders with 100% accuracy

Adverse events

None reported. That is not evidence of safety.

Monitoring: Not reported in the abstract.

The authors concluded (paraphrased): In a proof-of-concept analysis, the authors concluded that continuously collected wearable physiological data can predict functional status and response to elamipretide in patients with Barth syndrome.

Peptide Library’s assessment

An automated assessment, made by AI separately from the authors’ conclusions and not reviewed by a person. How risk of bias is assessed.

Limitations

small or imprecise: Only 12 patients were analysed. Accuracy estimates, including perfect 100% classification, are highly unstable and prone to overfitting.
inappropriate endpoint: Outcomes are clustering accuracies. The abstract reports no elamipretide-versus-placebo treatment effect on function, fatigue or safety.
selective reporting: Post hoc analysis with many models and median-split groupings. Only the highest accuracies are highlighted in the abstract.
abstract only: Only the abstract is available. Randomization, dosing, follow-up and responder-definition details cannot be checked.
other: Responder status was defined by median split of within-trial change, an arbitrary threshold not validated as clinically meaningful.

Accuracy of clustering patients by functional status from wearable physiological data: High risk of bias (Cochrane RoB 2)

Risk of bias in one result of a randomized trial, judged in five domains.

Domain judgments
Randomization process	Some concerns	The abstract states the trial was randomized and double-blind but gives no sequence generation or concealment details.
Deviations from intended interventions	Some concerns	No information on adherence, crossover carryover or deviations from intended treatment in the abstract.
Missing outcome data	Some concerns	Completeness of wearable time-series data and functional scores for all 12 patients is not reported.
Measurement of the outcome	Some concerns	Functional groups were formed by median split, and the clustering approach, validation and labelling are not described in detail.
Selection of the reported result	High risk of bias	Post hoc analysis with multiple scores and models, reporting a range of accuracies (60-93%) and highlighting the best ones.

Post hoc multi-model analysis in 12 patients with selective highlighting of top accuracies; reported-result domain is high risk.

Version 1 · published November 7, 2026 · evidence through November 7, 2026 · research snapshot 2026.1.1 · methodology ev-1 · RoB 2 (2019-08-22) · Automated assessment by AI (claude-opus-5-5 and claude-sonnet-5-5), not reviewed by a person.

Accuracy of clustering responders vs non-responders to elamipretide: High risk of bias (Cochrane RoB 2)

Risk of bias in one result of a randomized trial, judged in five domains.

Domain judgments
Randomization process	Some concerns	Randomization is stated but its methods are not described in the abstract-only record.
Deviations from intended interventions	Some concerns	Crossover conduct, washout and adherence are not reported in the abstract.
Missing outcome data	Some concerns	The abstract does not state whether all 12 patients contributed complete data to the response models.
Measurement of the outcome	Some concerns	Response was defined by median split of functional score change, an arbitrary, unvalidated responder definition.
Selection of the reported result	High risk of bias	Perfect 100% accuracy across all models in 12 patients suggests overfitting, and no external validation is reported.

Perfect accuracy from post hoc clustering in a tiny sample without validation carries high risk of bias in the reported result.

Claims this study informs

Improves function in Barth syndrome: not eligible (insufficient data) for “Muscle strength, exercise capacity and fatigue”
Improves exercise capacity in primary mitochondrial myopathy: not eligible (wrong population) for “Six-minute walk distance and fatigue”
Is safe for human administration: not eligible (insufficient data) for “Adverse events, including injection-site reactions, and discontinuations”

Where each value comes from

arms: PMID 37041653, abstract (Results, sentence 1) · extracted by AI (automated, not verified by a person) · abstract only
author_conclusion: PMID 37041653, abstract (Conclusions, sentence 1) · extracted by AI (automated, not verified by a person) · abstract only
design: PMID 37041653, abstract (Results, sentence 1) · extracted by AI (automated, not verified by a person) · abstract only
funding: PMID 37041653, abstract (PubMed publication types) · extracted by AI (automated, not verified by a person) · abstract only
limitations: PMID 37041653, abstract (Results, sentences 3-6) · extracted by AI (automated, not verified by a person) · abstract only
outcomes: PMID 37041653, abstract (Results, sentences 2-4) · extracted by AI (automated, not verified by a person) · abstract only
participants: PMID 37041653, abstract (Results, sentence 1) · extracted by AI (automated, not verified by a person) · abstract only
population: PMID 37041653, abstract (Background, sentences 1-3) · extracted by AI (automated, not verified by a person) · abstract only
results: PMID 37041653, abstract (Results, sentences 5-6) · extracted by AI (automated, not verified by a person) · abstract only

Version 1 · published November 7, 2026 · evidence through November 7, 2026 · research snapshot 2026.1.1 · methodology ev-1 · Automated assessment by AI (claude-opus-5-5 and claude-sonnet-5-5), not reviewed by a person.

These assessments describe published research. Records marked automated were produced by AI under the published methodology and have not been reviewed by a person; they can contain errors. None of this is medical advice, not a statement about any product’s safety for human use, and not a regulatory approval. Peptide Library’s vendor and partner relationships play no part in them (editorial independence).